THIS PAPER DESCRIBES A NEW SEGMENTATION METHOD FOR JAPANESE PRINTED DOCUMENTS USING LAYOUT KNOWLEDGE. MOST JAPANESE DOCUMENTS USE BOTH JAPANESE CHARACTERS AND ENGLISH WORDS. IT IS VERY DIFFICULT FOR A RECOGNITION MACHINE OF JAPANESE DOCUMENTS TO RECOGNIZE WHETHER THE CHARACTER STRINGS ARE ENGLISH WORDS OR JAPANESE WORDS. THE FIXED PITCH AND MIXED PITCH SEGMENTATION ALGORITHMS CAN EXTRACT ENGLISH WORDS FROM JAPANESE SENTENCES. IT CONCLUDES THAT A MIXED PITCH SEGMENTATION ALGORITHM FOR CHARACTER RECOGNITION IS IMPROVED.
